import pandas as pd
import matplotlib as plt

plt.rcParams['savefig.dpi'] = 900
plt.rcParams['figure.dpi'] = 900

data = pd.read_csv('Ecommerce data.csv')
data.head()

	id	logtarg	r	f	m	tof	Ffiction1	Fclassics3	Fcartoons5	Flegends6	...	Mhealth35	Mcooking36	Mlearning37	MGamesRiddles38	Msports39	Mhobby40	Mnature41	Mencyclopaedia44	Mvideos50	Mnonbooks99
0	914	NaN	194	7	318.894775	1703	1	0	1	0	...	14.899994	10.174706	0.000000	0.0	0.0	0.000000	0.000000	0.0	0.000000	0.0
1	957	NaN	3	14	368.052246	2364	1	0	2	0	...	74.156250	0.000000	22.749985	0.0	0.0	61.176025	17.150269	10.0	0.000000	0.0
2	1406	NaN	1489	15	423.298340	2371	0	0	0	0	...	0.000000	0.000000	0.000000	0.0	0.0	0.000000	0.000000	0.0	0.000000	0.0
3	1414	NaN	155	4	71.217041	1290	0	0	0	0	...	0.000000	0.000000	0.000000	0.0	0.0	0.000000	0.000000	0.0	0.000000	0.0
4	1546	NaN	194	6	442.638184	2188	0	0	1	0	...	41.518219	0.000000	7.949997	0.0	0.0	17.895203	0.000000	0.0	61.799988	0.0

5 rows × 66 columns

General Statistics of Variable r, f, m

data[["r", "f", "m"]].describe()

	r	f	m
count	33713.000000	33713.000000	33713.000000
mean	508.427254	5.820603	251.924325
std	542.440039	7.008049	3929.929826
min	0.000000	0.000000	0.000000
25%	117.000000	1.000000	39.899994
50%	293.000000	3.000000	102.629944
75%	712.000000	7.000000	247.297485
max	2460.000000	118.000000	532892.000000

import matplotlib.pyplot as plt
import seaborn as sns

plt.figure(figsize=(15, 5))

# Recency Histogram
plt.subplot(1, 3, 1)
sns.histplot(data['r'], bins=30, kde=False)
plt.title('Distribution of Recency')

# Frequency Histogram
plt.subplot(1, 3, 2)
sns.histplot(data['f'], bins=30, kde=False)
plt.title('Distribution of Frequency')

# Monetary Histogram
plt.subplot(1, 3, 3)
sns.histplot(data['m'], bins=30, kde=False)
plt.title('Distribution of Monetary Value')

plt.tight_layout()
plt.show()

Conclusion:

• Recency & Frequency: Right-skewed
• Monetary Value: Extremely right-skewed. Most customers have a relatively low total monetary value, and very few have a high monetary value.The long tail extending to the right indicates the presence of outliers with exceptionally high spend on the website. These are likely some VIP customers who contribute a significant portion of the revenue.

RFM Analysis

1. Recency and Frequency: Segment these using quartiles (0.25, 0.5, 0.75), with scores from 1 to 4. More recent and more frequent purchases get higher scores.

2. Monetary: Also use quartiles for segmentation, but additionally identify the extreme outliers. These customers will be handled separately as they represent a highly valuable segment. Scores from 1 to 5, where 5 stands for outliers.

3. Outlier Handling: Define a threshold for outliers in the "m" dimension. Here consider a fixed monetary value 100,000, depending on the business context.

4. Assigning Scores: After segmenting based on quartiles, assign a special score to indicate customer segments.

# Ensure numeric data types and handle missing values
data['r'] = pd.to_numeric(data['r'], errors='coerce')
data['f'] = pd.to_numeric(data['f'], errors='coerce')
data['m'] = pd.to_numeric(data['m'], errors='coerce')
data = data.dropna(subset=['r', 'f', 'm'])

# Define quartiles for R and F
r_labels = range(4, 0, -1)  # Since recent customers are more valuable, they get a higher score
f_labels = range(1, 5)
r_quartiles = pd.qcut(data['r'], q=4, labels=r_labels)
f_quartiles = pd.qcut(data['f'], q=4, labels=f_labels)

# Assign quartile labels to the data
data['R_Score'] = r_quartiles
data['F_Score'] = f_quartiles

# For 'm', first define a new category '5' for high spenders
data['M_Score'] = pd.qcut(data['m'], q=4, labels=False) + 1  # This will assign labels from 1 to 4
high_spender_threshold = 100000
data.loc[data['m'] > high_spender_threshold, 'M_Score'] = 5  # Assign '5' to high spenders

# Combine the RFM scores into a single score
data['RFM_Score'] = data['R_Score'].astype(str) + data['F_Score'].astype(str) + data['M_Score'].astype(str)

# Output the first few rows to confirm the RFM scoring
data[['r', 'f', 'm', 'R_Score', 'F_Score', 'M_Score', 'RFM_Score']]

	r	f	m	R_Score	F_Score	M_Score	RFM_Score
0	194	7	318.894775	3	3	4	334
1	3	14	368.052246	4	4	4	444
2	1489	15	423.298340	1	4	4	144
3	155	4	71.217041	3	3	2	332
4	194	6	442.638184	3	3	4	334
...	...	...	...	...	...	...	...
33708	6	1	0.000000	4	1	1	411
33709	4	1	0.000000	4	1	1	411
33710	5	1	22.949997	4	1	1	411
33711	0	0	0.000000	4	1	1	411
33712	2	1	4.949997	4	1	1	411

33713 rows × 7 columns

Segmentation Rule:

(I think I've guaranteed MECE. Apart from "5" in m_score, the other $ 4\times4\times4 = 64 $ scores will be allocated into only one category and no left out) (Also this is not a good way of showing MECE. I'll beautify this)

def assign_rfm_segments(row):
    r, f, m = row['R_Score'], row['F_Score'], row['M_Score']
    r, f, m = int(r), int(f), int(m)
    
    if m == 5:
        return 'VIP Customers'
    elif m == 4 and r in [2, 3, 4]:
        return 'Big Spenders'
    elif f == 4 and m in [1, 2, 3]:
        return 'Loyal Customers'
    elif r == 4 and f in [1, 2, 3] and m in [1, 2, 3]:
        return 'Recent Incoming Customers'
    elif r in [2, 3] and f in [2, 3] and m in [2, 3]:
        return 'Promising Customers'
    elif (r == 1 and f in [2, 3] and m in [2, 3]) or (r == 1 and m == 4):
        return 'At Risk'
    elif r in [2, 3] and f == 1 and m in [1, 2, 3]:
        return 'Hibernating'
    elif r in [1, 2, 3] and f in [2, 3] and m == 1:
        return 'Low Value Customers'
    elif r == 1 and f == 1 and m in [1, 2, 3]:
        return 'Almost Lost'
    else:
        return 'Other'

# Apply the function to the dataframe
data['Segment'] = data.apply(assign_rfm_segments, axis=1)

# Check the first few rows to see the segments
data[['R_Score', 'F_Score', 'M_Score', 'Segment']]

	R_Score	F_Score	M_Score	Segment
0	3	3	4	Big Spenders
1	4	4	4	Big Spenders
2	1	4	4	At Risk
3	3	3	2	Promising Customers
4	3	3	4	Big Spenders
...	...	...	...	...
33708	4	1	1	Recent Incoming Customers
33709	4	1	1	Recent Incoming Customers
33710	4	1	1	Recent Incoming Customers
33711	4	1	1	Recent Incoming Customers
33712	4	1	1	Recent Incoming Customers

33713 rows × 4 columns

data['Segment'].value_counts()

Segment
Big Spenders                 7209
Promising Customers          5976
At Risk                      5039
Hibernating                  4673
Recent Incoming Customers    4603
Almost Lost                  2564
Loyal Customers              1958
Low Value Customers          1687
VIP Customers                   4
Name: count, dtype: int64

import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import Axes3D
import pandas as pd
import numpy as np

# Assuming 'data' is your DataFrame with the RFM scores and segments

# Define colors for each segment for visual distinction
color_dict = {
    'Big Spenders': 'pink',
    'Promising Customers': 'blue',
    'Hibernating': 'green',
    'Recent Incoming Customers': 'orange',
    'At Risk': 'purple',
    'Almost Lost': 'brown',
    'Loyal Customers': 'red',
    'Low Value Customers': 'olive',
    'VIP Customers': 'cyan'
}

# Create a 3D scatter plot
fig = plt.figure(figsize=(10, 8))
ax = fig.add_subplot(111, projection='3d')

# Plot each segment with a different color
for segment in data['Segment'].unique():
    segment_data = data[data['Segment'] == segment]  # Use a temporary variable for the filtered data
    ax.scatter(segment_data['r'], segment_data['f'], segment_data['m'], 
               label=segment, s=20, alpha=0.6, color=color_dict[segment])

# Label the axes
ax.set_xlabel('Recency')
ax.set_ylabel('Frequency')
ax.set_zlabel('Monetary')

# Legend and title
plt.legend()
plt.title('RFM Segmentation (3D Visualization)')

# Show plot
plt.show()

Question: Is this a good way of showing segments? Or treemap is better?

Synthesize New Customers

import numpy as np

np.random.seed(42)

# Generate random data following an exponential distribution (because of the shape of distribution)
new_customers = pd.DataFrame({
    'CustomerID': range(1, 11),
    'r': np.random.exponential(scale=508, size=10).astype(int),
    'f': np.random.exponential(scale=4.82, size=10).astype(int) + 1,  # Frequency must be at least 1
    'm': np.random.exponential(scale=251, size=10)
})

new_customers['m'] = np.maximum(new_customers['m'], 10)

new_customers

	CustomerID	r	f	m
0	1	238	1	237.539089
1	2	1529	17	37.708867
2	3	668	9	86.724394
3	4	463	2	114.525582
4	5	86	1	152.842607
5	6	86	1	386.021939
6	7	30	2	55.906701
7	8	1021	4	181.229318
8	9	466	3	225.273685
9	10	625	2	11.938526

def assign_rfm_segments(row, r_quartiles, f_quartiles, m_quartiles):

    r_score = sum(row['r'] >= bound for bound in r_quartiles) + 1
    f_score = sum(row['f'] <= bound for bound in f_quartiles) + 1
    m_score = sum(row['m'] <= bound for bound in m_quartiles) + 1

    r, f, m = r_score, f_score, m_score
    
    if m == 5:
        return 'VIP Customers'
    elif m == 4 and r in [2, 3, 4]:
        return 'Big Spenders'
    elif f == 4 and m in [1, 2, 3]:
        return 'Loyal Customers'
    elif r == 4 and f in [1, 2, 3] and m in [1, 2, 3]:
        return 'Recent Incoming Customers'
    elif r in [2, 3] and f in [2, 3] and m in [2, 3]:
        return 'Promising Customers'
    elif (r == 1 and f in [2, 3] and m in [2, 3]) or (r == 1 and m == 4):
        return 'At Risk'
    elif r in [2, 3] and f == 1 and m in [1, 2, 3]:
        return 'Hibernating'
    elif r in [1, 2, 3] and f in [2, 3] and m == 1:
        return 'Low Value Customers'
    elif r == 1 and f == 1 and m in [1, 2, 3]:
        return 'Almost Lost'
    else:
        return 'Other'

r_quartiles = [117, 293, 712]  # Thresholds using quantiles
f_quartiles = [1, 3, 7]
m_quartiles = [39.9, 102.63, 247.3, 100000]

new_customers['Segment'] = new_customers.apply(
    assign_rfm_segments, axis=1, r_quartiles=r_quartiles, f_quartiles=f_quartiles, m_quartiles=m_quartiles)

new_customers[['CustomerID', 'r', 'f', 'm', 'Segment']]

	CustomerID	r	f	m	Segment
0	1	238	1	237.539089	Loyal Customers
1	2	1529	17	37.708867	VIP Customers
2	3	668	9	86.724394	Big Spenders
3	4	463	2	114.525582	Promising Customers
4	5	86	1	152.842607	Loyal Customers
5	6	86	1	386.021939	Loyal Customers
6	7	30	2	55.906701	At Risk
7	8	1021	4	181.229318	Recent Incoming Customers
8	9	466	3	225.273685	Promising Customers
9	10	625	2	11.938526	VIP Customers